Hello Tech Energy Adds a Claude 3 Sonnet and Bedrock Knowledge Base Plug-in to Its Overseas Customer Service

Shenzhen Hello Tech Energy, a portable energy storage company behind the Jackery brand, has been consolidating its overseas customer service onto one system since 2022. According to an AWS case study, the team faced rising ticket volumes, which reach 1.5 to 2 times normal levels during Black Friday and Prime Day, enquiries spread across phone, SMS, email, social media and Amazon marketplace messages, slow first responses and a thin internal knowledge base. The existing system could not add custom intent labels and could only generate content from its vendor's own knowledge. AWS extended the system with a plug-in built for Hello Tech's overseas business. For each ticket the plug-in calls Anthropic's Claude 3 Sonnet through Amazon Bedrock, matches the enquiry against an intent list set by the service managers, and classifies it as pre-sale, post-sale or repurchase, with a custom labelling scheme. Managers can configure different prompts per intent. Agents only upload knowledge documents; Knowledge Bases for Amazon Bedrock handles chunking, embedding and storage, supports multi-turn context and works with Bedrock Guardrails for content filtering. The stack also uses Amazon OpenSearch Serverless and Amazon DynamoDB. When a query arrives, the plug-in retrieves knowledge by intent and drafts a reply. Agents mark each draft as usable, unsuitable or needing work, and the feedback is stored so that staff responsible for the knowledge base can find gaps or outdated content and fill them. The plug-in went live on 19 April and went through three versions. By around 22 May agents were sending generated content directly for more than 30% of enquiries, with a target of about 40%. AWS reports a 70% average first-contact resolution rate, costs of 1/20 of the previous third-party AI tool, and, for the pilot alone, time savings equivalent to two agents a year who can be moved to customer insight work. Liu Zhiyuan, IT product manager for the intelligent call centre, said the company will keep testing generative AI through low-cost trials and fast validation.